On apparently non-modal evidentials

Current literature offers a range of analyses of evidentials in natural language, which can be broadly grouped into two types: modal analyses (Kratzer 1991, Izvorski 1997, Ehrich 2001, Garrett 2001, Faller 2006, Matthewson et al. 2007, McCready and Asher 2006, McCready and Ogata 2007, Waldie et al. 2009, Peterson 2009, 2010, Lee this volume, among others), and non-modal analyses (Faller 2002, 2003, Chung 2005, Portner 2006, Davis et al. 2007, Murray 2009a,b, Peterson 2009, 2010, among others). The split between modal and non-modal analyses correlates with significant empirical differences between the groups of evidentials being analyzed (as outlined in section 2 below). However, the question arises of what distinguishes the many distinct non-modal approaches from each other. The goal of this paper is to test available non-modal analyses against one previously unanalyzed evidential in St’át’imcets (Lillooet Salish): lákw7a. I will argue that with respect to all the usual diagnostic tests (including known truth or falsity of the prejacent proposition, the impossibility of canceling or explicitly denying the evidence source, and so on), the available non-modal approaches do not make different empirical predictions from each other, and all appear to be applicable to lákw7a. However, I then show that lákw7a poses a problem for all non-modal analyses. In order to account for the evidence source restriction of lákw7a, we need to adopt Faller’s (2003) notion of non-overlap between the event-trace and the speaker’s perceptual field. If this is correct, then lákw7a must operate at the event level; this in turnmeans that it cannot be captured by any non-modal analyses, as all of these entail that evidentials operate at a level distinct from the propositional content. I conclude by arguing that lákw7a is a modal evidential after all. Following Matthewson 2009, 2010 (which in turn relies on Kratzer 2010, von Fintel and Gillies 2010), I suggest that the apparently significant empirical differences between the two classes of evidentials do not force us to abandon a modal analysis for any evidential.